The Mystery Behind the Globs of Epoxy

When Sparkfun visited the factory that makes their multimeters and photographed a mysterious industrial process.

We all know that the little black globs on electronics has a semiconductor of some sort hiding beneath, but the process is one that’s not really explored much in the home shop.  The basic story being that, for various reasons , there is no cheaper way to get a chip on a board than to use the aptly named chip-on-board or COB process. Without the expense of encapsulating  the raw chunk of etched and plated silicon, the semiconductor retailer can sell the chip for pennies. It’s also a great way to accept delivery of custom silicon or place a grouping of chips closely together while maintaining a cheap, reliable, and low-profile package.

As SparkFun reveals, the story begins with a tray of silicon wafers. A person epoxies the wafer with some conductive glue to its place on the board. Surprisingly, alignment isn’t critical. The epoxy dries and then the circuit board is taken to a, “semi-automatic thermosonic wire bonding machine,” and slotted into a fixture at its base. The awesomely named machine needs the operator to find the center of the first two pads to be bonded with wire. Using this information it quickly bonds the pads on the silicon wafer to the  board — a process you’ll find satisfying in the clip below.

The final step is to place the familiar black blob of epoxy over the assembly and bake the board at the temperature the recipe in the datasheet demands. It’s a common manufacturing process that saves more money than coloring a multimeter anything other than yellow.

67 thoughts on “The Mystery Behind the Globs of Epoxy

        1. i think it’s just pressed against the copper. Maybe is just deforming the copper enough to solder it to the copper.
          There are gold wires and copper wires, i guess they’re pure, becuase pure copper and pure gold have more conductivity.

          1. Au is like butter, as in I had a problem once carrying some Canadian gold coins in a back pocket and when I reached the destination some had been bent. I think this is why in the old movies they showed the old prospector biting gold coins, a Jolly Rancher candy is harder than un-alloyed gold.

          2. It’s a weld performed using an ultrasonic transducer. Gold is one metal but IIRC, there can be issues with gold and some other metals. I think copper wire would need an inert gas but it’s been a long time since I learned about this stuff.

  1. It makes me wonder why is it even viable to do so. There are multiple steps in the process, each taking time. Chip packaging and pick-and-place are so automated that even without considering cheap labor cost, you save time and defects.
    Now, if it were a bigger chip I would understand, but this is fairly low pin count.

      1. Very good point. There’s no question but that this technique provides a good level of security, but economics of manufacture is, quite probably, still be the overriding reason for its use
        Just a guess.
        (And I think it’s a bummer that inexpensive YELLOW multimeters are no longer available. They stand out so well in all my tool boxes and bags. I simply won’t buy that other brand)

    1. I suspect it’s viable in this case only because the chip is fairly-custom and used in small quantities, so the labour costs are no worse than having the chip manually packaged.

      For super-cheap things like musical cards and LCD controllers, the quantities are huge and I’d bet that the process has no labour in it whatsoever. The bond-wire placement would be automated in exactly the same way that it is when packaging the silicon in LQFP or whatever, and a step is saved.

      1. That and the relatively low numbers of a custom chip must be the reason why this is cheaper than packaging the die and just having the pick and place process handle it. I was wondering in what way what is shown could be advantageous.

    2. Wires need to be bonded to the chip somewhere down the line. Normally the chip manufacturer does it, and bonds it to the pins on the chip’s casing. If you relieve them of that responsibility, you get the chips cheaper. The price difference is apparently enough to make it worth doing yourself. If you’re a big enough factory, that is, I imagine this multimeter factory makes hundreds of thousands or millions of them.

      Doing it this way saves on having to connect a plastic-cased chip’s pins to the board, so that’s a step removed.

  2. Did anybody else catch that bullshit “wireless anti-static strap” in the video.

    You see these a lot in Chinese factories manufacturing electronics. Which is strange in my opinion since they are all made in China anyhow. It would cost maybe a penny more per person for them to have proper ESD grounding in their factory.

    1. As I see it, the “wireless anti-static strap” is meant for other workstations than the one he is sitting at, at the moment. As he’s wearing an anti static glove on the hand that comes in contact with the PCB. Same principle we are using at the company I work at.
      – Of course when he touches the little leaver that locks the PCB in place with his unprotected hand, the idea behind it kind of goes out the window.

    2. > Did anybody else catch that bullshit “wireless anti-static strap”

      Having a sufficiently strong source of ionizing radiation should render air conductive enough that it might work.

      I’d recommend a couple of tens of GBq [1], i.e. around one gram of radium. Portable, clean and… wireless.

      (Yeah, I jest, I jest).


      1. It is not just a question of UV light. Silicon junctions are light sensitive down to the NIR range, to 1100nm. So as it’s sensitivity spectrum is wider than the human eye, there is only one color to protect against the whole range: black.

      1. Which would not be a problem. Everything is clearly marked so this IC is the well known ICL7106 or china clone of it. I checked with the datasheet of the ICL7106. So yes, you could grind of the blob and replace it with the MQFP version. But I am not sure, if this would be economic.

    1. Yeah but cheaper to buy. Depending on the failure rate, it’s presumably cheaper, considering all customers, to make them unrepairable but cheaper, than repairable and more expensive. Per 100,000 customers it’s probably much cheaper to offer a replacement guarantee and just throw away the broken ones.

      Things aren’t made for repair any more, it’s economics. Partly down to the cheap price of components compared to human labour, partly down to increased reliability.

    1. I’m feeding a troll I’m sure. Not impossible, just difficult. You could cut the traces to the OBC to have points to connect your home bench constructed circuitry that mimics the function of the OBC

    1. You can do it by hand, but it does require some care to do well, more than most soldering work. If your bonds fail you will ruin the chip/die, so probably more cost effective to have it automatic and much faster. I can do a bond every maybe 10 seconds, the machines obviously much faster.

      1. Is it something you can do at home or does it require a $5000 piece of equipment to do it manually?
        I haven’t checked on the availability of chips in home shop quantities but it seems like you could avoid stocking multiple package types and decrease board size.

  3. This reminds me it has been quite some time I haven’t seen the 99 cent wrist watches being sold for some time now. I guess the consumer gave up that chees for the still cheesy >$10 metal case and band wrist watches. Nuy hey I’m not above using functional chees. I suppose hacker will grind that Fluke Vs. Sparkfun axe for decades. While the Fluke appearance trade mark was broad and vague, Sparkfun should have known better that to poke at a giant.

  4. Gold wire can lead to trouble that doesn’t involve electronics. Years ago a person who worked at the Micron plant in Mountain Home, Idaho stole a spool of gold bonding wire.

    So far, so good for the theft. Then he went to Boise and tried to pawn it, with the Micron label on the spool. Of course he got busted.

    1. You probably don’t have to give an exact position, more like “the fiducial is in corner”. Speaking from experience with an LPKF mill, automatically registering a fiducial is easy. Determining that you registered the wrong one less so ;)

  5. In the 80’s I found an article in a 1970’s electronics magazine on how to add constant and memory functions to a specific make and model of handheld calculator.

    The picture looked familiar, digging around in my stuff turned up a calculator that looked identical to the one in the magazine. Same brand, same number etc. I-den-ticle! Then I opened it up. Instead of the board with separate chip and other components like in the article, there were some unused screw posts, the back side of the key pad and a chunk of ribbon cable up to the LED display board with a single epoxy blob on its back. :(

    Somewhere along the line the manufacturer did a drastic simplification on that model, stripping it down to the basic 4-banger the original release had only pretended to be.

    1. So the original calculator had the extra functions all along, they were just disabled in the cheaper model? To think how much of a fuss that generated back when Intel started doing it with their CPUs, and more lately, Tek with their oscilloscopes (I think it was Tek).

      Still I knew it wasn’t a modern invention. There was an upgrade to one of the IBM mainframes, way back in the day, late ’60s early ’70s I think. Doubled the speed. Involved removing a jumper, or maybe it was fitting one. An IBM guy would come out and clear the room, then fit the special jumper, doubtless among a lot of theatrics for the customer’s benefit. Remembering to take the jumper back out when the IBM guy came round was an important responsibility, for shops who figured the “upgrade” out for themselves.

      This was a while after the earlier leased computers, which had a car-style odometer that counted how many instructions you’d executed. IBM would then bill you by the instruction, or the kilo-instruction at least. Who knew computers were once expensive enough to meter?

      On the other note, I’ve seen books from back in the 1970s, telling people how to add a calculator chip to your home-brew computer system. The type of computer that would have an 8-bit CPU and a couple of K RAM. You could add a calculator chip, to give it the power of an FPU! The method was to connect the calculator chip’s keypad to your computer’s bus with appropriate glue logic, and the same for it’s display. Then operate the calculator chip as if it was by hand. I really can’t imagine how that’d be quicker than just doing the calculation yourself in software on the CPU. But it at least made it into articles, whether or not anyone did it.

      Thing is, the calculator chip itself would generally have been a low-powered 4-bit CPU, running calculator software off it’s internal ROM. The chip had helper hardware on board to drive the 7-segment displays from simple BCD. I once had an old LED calculator, well, a couple, that took a second or more to do complicated sums.

      But yep, it was apparently a thing people did. Even though it seems completely counter-intuitive. Any good reason why it would have helped, that I’ve missed?

      1. Oh no, it was Rigol, not Tektronix. Search “Rigol” on this site’s search and you’ll find some examples. Sorry!

        Still would like to hear more of what you know on that old calculator upgrade though. The nature of the upgrade.

        Another little calc-related niggle is…

        A particular Casio organiser. They made others with the same LCD technology and even a couple of wristwatches IIRC.

        The LCD in question has 3 colours, orange-red, greenish, and blue. And clear, of course. What’s interesting is it’s not RGB. A single pixel changes colour, between those colours. The actual LCD cell changes what part of the spectrum it reflects. It’s related, apparently, to how other LCDs do brightness levels. Again, the type of LCD I mean is the text-display type, or the lo-res “graphical” ones you find in calculators. Not a modern video display, no RGB here.

        Instead of brightness, somehow it does colours instead. I believe it’s driven in the same way as variable brightness displays. The technology goes back to the early 1990s, took off for a while, then disappeared. Seems mostly a Casio speciality. I once had a bookmark for a company that made them in custom display formats, and I’m DYING to know how it works! The name, in particular. Something like “pseudo-colour”, except searching that came up with the same 200 Alibaba links to “Koycera” displays, of the standard RGB type and Kyocera not spelled properly.

        Damn cut and pasting Chinese. If it weren’t for the low margins and artificially low exchange rates, I’d start a business offering to grammar-check Chinese sellers’ spelling and grammar. I’d never run out of work. Unfortunately I don’t think the low cost of living and required command of English occur anywhere within a single country, or certainly not within a single person.

        In other Chinglish news, I remember a Bluetooth earpiece advertised as “stereo”. Just the one, single earpiece. For one ear.

          1. Yup it’s definitely CCSTN, color-coded super-twisted nematic.

            Question 2 is, I’d love to play with one of these, but no idea where to get one. Since I’m not an enormous industry who wants 500 of them. I appreciate it’s a bit of a niche, and I think unfortunately too much of a niche. I’d just like a graphical panel with these that I could play with.

            One manufacturer offers a range of 32 colours, of which you can pick 4 to go on your display. Dunno if you can pick any 4 of 32, or if they have to be related in some way, perhaps close enough, or far apart enough, from each other. Real shame, and having all 32 at once would be great! I suspected the Game Boy Color used one of these, but apparently not.

            Anyone knows a place who sell the modules in small quantities, I’d like to know. None of the obvious places do them. I suppose there’s not much demand, it’s either full RGB video-capable, or monochrome old-fashioned slow LCD. I want COLOUR old-fashioned slow LCD!

            Just the colours themselves are fascinating, something like a butterfly’s wing, but with a muddy, low-saturation quality. Seems like they’re produced by refraction, I bet looking from the wrong angle would make the colours change. Partly it’s nice because it’s an alternative to RGB that doesn’t seem to have gone anywhere, looks a bit obsolete.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s